How to use decode_predictions() for non-Imagenet models..?

Question

I know that decode_predictions() works for only imagenet dataset(1000 classes) for the models like VGG16 etc. But condiser my scenario.

My Scenario:

I used vgg16 pretrained model, and added my own weights.

So this turns out to be a non-Imagenet model. I have mentioned classes=9 as i trained my previous model with 9 classes only.

So now to find the predictions, i could use predict() method and then my print(answer) would give the corresponding class label. But actually i need the class name to be printed. Is that possible to get class name ?? If it is, can anyone explain me how.?

Answer 1

In Deep learning when you are performing prediction you will get prediction in your case it is an array with 9 probabilities in them. So first perform following operation on them.

import numpy as np

prediction = model.prediction(test_data) 
# prediction will contain [[0.6, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05]]

prediction = np.argmax(prediction[0])
# Now predition hold the index of the (0.6) i.e max probability value

Now after that you should have a dictionary which contains key as 0, 1, 2, .. 8 and values as classname1, classname2, ...classname9

This is the way you will get the class name as output